Network fault rapid recovery method and device, equipment and storage medium

Abstract

The invention provides a network fault rapid recovery method and device, equipment and a storage medium, and the method comprises the steps: selecting a target topological structure matched with a current time point from prediction topological structures corresponding to all prediction time points from a target network if the current network fault of the target network is detected; the prediction topological structure is predicted by an LSTM model in advance; extracting resource allocation strategy data of the target topological structure; the resource allocation strategy data is obtained through prediction of an SAT model in advance; the node feature vector is obtained through prediction of a GAT model in advance; and adjusting target network resource allocation according to the resource allocation strategy data of the target topological structure to realize network fault recovery. According to the method, the response speed of network fault recovery can be effectively improved, resource allocation is optimized, resource waste is reduced, and the automation level is improved through intelligent decision, so that the problems of slow response, unreasonable resource allocation and lack of intelligence in traditional network fault recovery are solved.

Description

Technical Field

[0001] This application relates to the field of communication technologies, and in particular, to a method, apparatus, device, and storage medium for rapid network fault recovery. Background Art

[0002] With the rapid development of information technology and the increasing complexity of network applications, network faults have become a key factor affecting communication quality and service stability. Most traditional network fault recovery methods are based on preset rules and manual intervention, often suffering from problems such as slow response speed, unreasonable resource allocation, and lack of intelligence, resulting in long network recovery time, low efficiency, and difficulty in adapting to the increasingly complex and dynamically changing network environment. Therefore, how to improve the intelligence and efficiency of network fault recovery has become an important topic in current research and practice. In the wave of digital transformation, the information systems of enterprises have become increasingly complex, which not only improves production efficiency and service quality but also increases the risk of system failures.

[0003] Traditional network fault recovery methods often adopt a one-size-fits-all approach, that is, all nodes recover in the same way and at the same speed after a fault occurs. However, this approach ignores the differences between different nodes, resulting in unreasonable resource allocation, slow recovery of important services, and a large amount of resources being occupied by the recovery of non-critical nodes. Currently, existing network fault recovery methods usually, after determining the occurrence of a network fault, use a shortest path algorithm (Dijkstra algorithm) or a heuristic search algorithm to find the shortest path from the starting point to the end point in the network, and then network fault recovery can be performed according to this path. However, since existing network fault recovery methods need to execute the network path calculation process after determining the occurrence of a network fault, it will lead to an extended time for network fault recovery. At the same time, for complex network topologies, existing network fault recovery methods also have problems such as high computational complexity and large memory consumption, which will further affect the response speed of network fault recovery. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method, apparatus, device, and storage medium for rapid network fault recovery to eliminate or improve one or more defects existing in the prior art.

[0005] One aspect of this application provides a method for rapid network fault recovery, including:

[0006] If it is detected that the target network currently has a network failure, then among the predicted topological structures corresponding to each prediction time point of the target network pre-acquired, select the predicted topological structure corresponding to one of the prediction time points that matches the current time point as the target topological structure; wherein, the predicted topological structures corresponding to each prediction time point are pre-predicted by the LSTM model according to the historical network state feature data and service feature data of the target network;

[0007] Extract the resource allocation policy data corresponding to the target topological structure from the resource allocation policy data corresponding to each of the pre-stored predicted topological structures; wherein, the resource allocation policy data corresponding to each of the predicted topological structures are pre-predicted by the SAT model according to the node feature vectors corresponding to each of the predicted topological structures and the service feature data; the node feature vectors corresponding to each of the predicted topological structures are pre-predicted by the GAT model according to the predicted topological structures corresponding to each prediction time point respectively;

[0008] Adjust the network resource allocation of the target network according to the resource allocation policy data corresponding to the target topological structure to achieve network failure recovery.

[0009] In some embodiments of the present application, before the step of if it is detected that the target network currently has a network failure, then among the predicted topological structures corresponding to each prediction time point of the target network pre-acquired, select the predicted topological structure corresponding to one of the prediction time points that matches the current time point as the target topological structure, it further includes:

[0010] Input the currently acquired historical network state feature data and service feature data of the target network into the LSTM model, so that the LSTM model outputs the predicted topological structures corresponding to each prediction time point of the target network respectively;

[0011] Store the correspondence between each prediction time point and the predicted topological structure corresponding to each prediction time point locally.

[0012] In some embodiments of the present application, before the step of if it is detected that the target network currently has a network failure, then among the predicted topological structures corresponding to each prediction time point of the target network pre-acquired, select the predicted topological structure corresponding to one of the prediction time points that matches the current time point as the target topological structure, it further includes:

[0013] Input the predicted topology structures and service feature data corresponding to each prediction time point of the target network output by the LSTM model into the GAT model, so that the GAT model correspondingly outputs the node feature vectors corresponding to the predicted topology structures at each of the prediction time points.

[0014] In some embodiments of the present application, before selecting, from the predicted topology structures corresponding to the target network at each prediction time point pre-acquired, the predicted topology structure corresponding to a prediction time point that matches the current time point as the target topology structure if it is detected that the target network currently has a network failure, it further includes:

[0015] Input the node feature vectors corresponding to the predicted topology structures output by the GAT model and the service feature data into the SAT model, so that the SAT model correspondingly outputs the resource allocation strategy data corresponding to the predicted topology structures;

[0016] Store the corresponding relationship between the predicted topology structures corresponding to each prediction time point and the resource allocation strategy data corresponding to each predicted topology structure locally.

[0017] In some embodiments of the present application, the inputting the historical network state feature data and service feature data of the currently acquired target network into the LSTM model includes:

[0018] Real-time acquire the network topology state data and service data corresponding to the target network at each historical time point respectively; wherein, the network topology state data includes the node state data corresponding to each node in the target network and the link state data corresponding to the links between each node; the service data includes service level, service traffic, delay requirement data, and bandwidth demand data;

[0019] Perform data cleaning on the network topology state data and service data corresponding to each historical time point respectively;

[0020] Extract features from the network topology state data corresponding to each historical time point after data cleaning to obtain the corresponding historical network state feature data; and extract features from the service data after data cleaning to obtain the corresponding service feature data;

[0021] Perform standardization and normalization processing on the historical network state feature data and service feature data.

[0022] In some embodiments of the present application, the adjusting the network resource allocation of the target network according to the resource allocation strategy data corresponding to the target topology structure to achieve network failure recovery includes:

[0023] According to the resource allocation policy data corresponding to the target topology structure, perform network topology adjustment, routing reconfiguration, and node load balancing processing on the target network.

[0024] In some embodiments of the present application, the adjusting the network resource allocation of the target network according to the resource allocation policy data corresponding to the target topology structure to achieve network fault recovery further includes:

[0025] During the network fault recovery process of the target network, if it is detected that the predicted topology structure corresponding to the predicted time point matching the current latest time point output by the LSTM model in advance has changed compared with the target topology structure, then update the current target topology structure to the predicted topology structure corresponding to the predicted time point matching the current latest time point;

[0026] Obtain the resource allocation policy data corresponding to the updated target topology structure, and re-perform network topology adjustment, routing reconfiguration, and node load balancing processing on the target network based on the resource allocation policy data corresponding to the updated target topology structure.

[0027] Another aspect of the present application provides a network fault rapid recovery device, including:

[0028] A predicted topology matching module, configured to, if it is detected that the target network currently has a network fault, select, from the predicted topology structures corresponding to the target network at each predicted time point pre-obtained, the predicted topology structure corresponding to one predicted time point matching the current time point as the target topology structure; wherein, the predicted topology structures corresponding to each predicted time point are pre-predicted by the LSTM model according to the historical network state feature data and service feature data of the target network;

[0029] An allocation policy extraction module, configured to extract the resource allocation policy data corresponding to the target topology structure from the resource allocation policy data corresponding to each of the pre-stored predicted topology structures; wherein, the resource allocation policy data corresponding to each of the predicted topology structures are pre-predicted by the SAT model according to the node feature vectors corresponding to each of the predicted topology structures and the service feature data; the node feature vectors corresponding to each of the predicted topology structures are pre-predicted by the GAT model according to the predicted topology structures corresponding to each predicted time point respectively;

[0030] A fault recovery module, configured to adjust the network resource allocation of the target network according to the resource allocation policy data corresponding to the target topology structure to achieve network fault recovery.

[0031] The third aspect of the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the network fault rapid recovery method described above is implemented.

[0032] The fourth aspect of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the network fault rapid recovery method described above is implemented.

[0033] The fifth aspect of the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, the network fault rapid recovery method described above is implemented.

[0034] For the network fault rapid recovery method provided by the present application, if it is detected that a network fault occurs in the target network currently, then among the predicted topological structures corresponding to each predicted time point of the target network pre-acquired, a predicted topological structure corresponding to a predicted time point that matches the current time point is selected as the target topological structure; wherein, the predicted topological structures corresponding to each predicted time point are respectively predicted in advance by an LSTM model according to the historical network state feature data and service feature data of the target network; from the resource allocation strategy data corresponding to each of the predicted topological structures pre-stored, the resource allocation strategy data corresponding to the target topological structure is extracted; wherein, the resource allocation strategy data corresponding to each of the predicted topological structures is respectively predicted in advance by an SAT model according to the node feature vectors corresponding to each of the predicted topological structures and the service feature data; the node feature vectors corresponding to each of the predicted topological structures are respectively predicted in advance by a GAT model according to the predicted topological structures corresponding to each predicted time point; the network resource allocation of the target network is adjusted according to the resource allocation strategy data corresponding to the target topological structure to achieve network fault recovery, which can effectively improve the response speed of network fault recovery, optimize resource allocation, reduce resource waste, and improve the automation level through intelligent decision-making, thereby solving the problems of slow response, unreasonable resource allocation, and lack of intelligence existing in traditional network fault recovery.

[0035] The additional advantages, objectives, and features of the present application will be partially described below, and will become partially obvious to those of ordinary skill in the art after studying the following text, or can be learned from the practice of the present application. The objectives and other advantages of the present application can be achieved and obtained through the structures specifically pointed out in the description and the drawings.

[0036] Those skilled in the art will understand that the objectives and advantages achievable with the present application are not limited to those specifically described above, and the above and other objectives achievable with the present application will be more clearly understood from the following detailed description. Description of the Drawings

[0037] The drawings described herein are used to provide a further understanding of the present application, form a part of the present application, and do not limit the present application. The components in the drawings are not drawn to scale, but are only for showing the principles of the present application. For the convenience of showing and describing some parts of the present application, the corresponding parts in the drawings may be enlarged, that is, may become larger relative to other components in the exemplary device actually manufactured according to the present application. In the drawings:

[0038] Figure 1 It is the first flowchart of the network fault rapid recovery method in an embodiment of the present application.

[0039] Figure 2 It is the second flowchart of the network fault rapid recovery method in an embodiment of the present application.

[0040] Figure 3 It is the flowchart of the network fault rapid recovery method in the video in the application example of the present application.

[0041] Figure 4 It is the schematic diagram of the LSTM model in the video in the application example of the present application.

[0042] Figure 5 It is the schematic diagram of the LSTM symbol meaning in the video in the application example of the present application.

[0043] Figure 6 It is the schematic diagram of the cell state of the LSTM in the video in the application example of the present application.

[0044] Figure 7 It is the schematic diagram of the LSTM forget gate in the video in the application example of the present application.

[0045] Figure 8 It is the schematic diagram of the LSTM state update value in the video in the application example of the present application.

[0046] Figure 9 It is the schematic diagram of the usage method of the LSTM input gate in the video in the application example of the present application.

[0047] Figure 10 It is the schematic diagram of the LSTM output gate in the video in the application example of the present application.

[0048] Figure 11 It is the schematic diagram of the GAT model in the video in the application example of the present application.

[0049] Figure 12 This is a schematic structural diagram of a network fault quick recovery device in an embodiment of the present application. Specific embodiments

[0050] To make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below in conjunction with the embodiments and the accompanying drawings. Herein, the illustrative embodiments of the present application and their descriptions are used to explain the present application, but not to limit the present application.

[0051] Herein, it should also be noted that in order to avoid obscuring the present application due to unnecessary details, only the structures and / or processing steps closely related to the solution of the present application are shown in the drawings, while other details less related to the present application are omitted.

[0052] It should be emphasized that the term "including / containing" when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0053] Herein, it should also be noted that if not otherwise specified, the term "connection" in this document can not only refer to a direct connection, but also represent an indirect connection with an intermediate.

[0054] In the following, embodiments of the present application will be described with reference to the drawings. In the drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0055] It should be noted that the shortest path algorithm (Dijkstra's algorithm) is one of the most classic algorithms in network routing algorithms. It is used to find the shortest paths from a source node to all other nodes. It uses a greedy strategy and gradually expands the known shortest paths. The shortest path algorithm has extensive applications in network topologies, especially in fields such as communication networks, transportation networks, and social networks. In communication networks, the shortest path algorithm is used to determine the transmission paths of data packets in the network. Taking Dijkstra's algorithm as an example, it can effectively calculate the shortest paths from the source node to all other nodes, ensuring that data packets are transmitted with the minimum delay. This is particularly important in the network management of large Internet service providers, which can improve bandwidth utilization and network response speed. In traffic navigation applications (such as Google Maps and Baidu Maps), the shortest path algorithm is used to provide users with the best driving routes. The system will calculate the time and distance of different paths in real time and select the optimal route, thereby improving the travel efficiency of users. This not only reduces travel time but also alleviates traffic congestion. In social networks, the shortest path algorithm helps analyze the connection relationships between users. For example, platforms such as Facebook and LinkedIn use this algorithm to recommend friends or potential business contacts. By calculating the shortest paths between users, key nodes and potential influencers in the social network can be identified, thus optimizing the user experience.

[0056] Although the shortest path algorithm has achieved remarkable results in multiple fields, there are also some technical drawbacks: 1. Computational complexity: For large networks, especially graph structures with a large number of nodes and edges, the computational complexity of the shortest path algorithm may lead to long calculation times, especially in dynamic environments where paths change frequently, making real-time calculation difficult. 2. Static assumptions: Most shortest path algorithms (such as Dijkstra's algorithm) are based on static network assumptions and cannot effectively handle dynamic changes in network states, such as node failures or link interruptions. This may require frequent recalculations in practical applications, affecting real-time performance. 3. Inapplicability to bandwidth limitations: The shortest path algorithm usually only focuses on path length (such as distance or time) and ignores other important factors such as bandwidth. In some cases, there may be bandwidth bottlenecks in the network, resulting in the shortest path not being the best path.

[0057] In summary, although the application of the shortest path algorithm in network topologies brings significant effects, the technical drawbacks cannot be ignored, and these factors need to be comprehensively considered in practical applications.

[0058] In addition, the A* algorithm is a commonly used heuristic search algorithm mainly used for graph path planning. The A* algorithm combines the advantages of the Dijkstra algorithm and heuristic search, aiming to find the shortest path from the starting point to the ending point in a graph structure. It determines the search priority by evaluating the "cost" of each node. The heuristic function is the core of the A* algorithm, which determines the efficiency of the algorithm and the quality of the path. Common heuristic functions include: Manhattan distance (suitable for grid-like graphs) and Euclidean distance (suitable for spatial graphs). The A* algorithm has a wide range of applications in network topologies, especially in fields such as path planning, routing selection, and game development. For example: 1. Network routing selection: In computer networks, the A* algorithm is often used for dynamic routing selection. For example, in large networks, routers need to select the optimal path to reduce latency and congestion. The A algorithm effectively finds the best route by considering the latency of each path and the estimated latency to reach the target. 2. Traffic navigation systems: Many navigation applications, such as Google Maps and Waze, use the A algorithm for path planning. These systems not only consider road distances but also factors such as real-time traffic conditions and section restrictions. By dynamically updating the heuristic function, the A algorithm can adjust the recommended path in real time and optimize the user travel experience. 3. Game development: In video games, the A* algorithm is widely used for the movement path planning of non-player characters (NPCs). NPCs need to find the shortest path in a complex environment to track players or avoid obstacles. The A algorithm can calculate the best path in real time according to the changes in the environment, making the game more realistic and interactive.

[0059] Although the A* algorithm has many advantages, there are also some technical drawbacks in practical applications: 1. Memory consumption: The A* algorithm needs to maintain an open list and a closed list during operation. In large networks or complex scenarios, storing information about a large number of nodes will cause significant memory consumption, which may limit the application scope of the algorithm. Selection of the heuristic function: The design of the heuristic function is crucial for the performance of the A* algorithm. If the heuristic function is not reasonable enough, it may lead to low search efficiency and even fail to find the optimal path in some cases. Designing a heuristic function that is both accurate and efficient is a challenge. 2. Adaptability to dynamic environments: In dynamic environments, the network topology may change, such as node failures or path blockages. Although the A* algorithm can handle these changes to a certain extent, frequent path recalculation will increase the computational burden and affect real-time performance. 3. Computational complexity: In some cases, especially when the complexity of the graph structure is high, the time complexity of the A* algorithm may reach O(b^d), where b is the branching factor of each node and d is the depth of the optimal solution. This means that in specific cases, the efficiency of the A* algorithm will be significantly affected.

[0060] Based on this, in order to improve the response speed of network fault recovery, optimize resource allocation, and reduce resource waste, the embodiments of the present application respectively provide a network fault rapid recovery method, a network fault rapid recovery device for executing the network fault rapid recovery method, an electronic device, a computer-readable storage medium, and a computer program product, which can effectively improve the response speed of network fault recovery, optimize resource allocation, reduce resource waste, and improve the automation level through intelligent decision-making, thereby solving the problems of slow response, unreasonable resource allocation, and lack of intelligence in traditional network fault recovery.

[0061] Specifically, it will be described in detail through the following embodiments.

[0062] Based on this, the embodiments of the present application provide a network fault rapid recovery method that can be implemented by a network fault rapid recovery device. Refer to Figure 1 , and the network fault rapid recovery method specifically includes the following content:

[0063] Step 100: If it is detected that a network fault occurs in the target network currently, then from the predicted topology structures corresponding to each predicted time point of the target network pre-obtained, select the predicted topology structure corresponding to one of the predicted time points that matches the current time point as the target topology structure; wherein, the predicted topology structures corresponding to each predicted time point are pre-predicted by an LSTM model according to the historical network state feature data and service feature data of the target network.

[0064] In one or more embodiments of the present application, the predicted topology structure includes network topology feature data corresponding to each predicted time point, and may also include data such as node status and link status; the service feature data includes feature data such as service level, service traffic, latency requirement, and bandwidth demand.

[0065] It can be understood that the predicted topological structure at the predicted time point refers to the predicted topological structure at a future time point predicted by the LSTM model at the time point when it performs prediction. That is to say, assuming that the respective historical time points corresponding to the historical network state feature data of the target network are t11, t12, t13…, t1n, where n is a positive integer greater than 3; the time point when the LSTM model performs prediction is t2, and the respective predicted time points to which the 5 predicted topological structures predicted by the LSTM model belong are t3, t4, t5, t6, and t7 respectively. Among them, t11, t12, t13…, t1n increase in sequence, and t1n, t2 to t7 increase in sequence. The current time point is any one of t3 to t7. Correspondingly, in order to quickly and accurately match the predicted topological structure corresponding to the predicted time point when it is detected that the target network currently has a network failure, the network failure fast recovery method needs to execute the LSTM model and the subsequent GAT and SAT processes in real time, and requires that the time interval between t1n and t3 is appropriate, such as an interval of 5S to 10min, etc., so that the current time point can match the predicted time point. Here, the match can mean exactly the same, or it can mean that the time difference between the two does not exceed a preset threshold, such as 1S to 1min, etc.

[0066] It can be understood that GAT is a graph attention network, which is a graph neural network based on the spatial domain; SAT is a self-attention mechanism model; LSTM is a long short-term memory network, which is a deep learning model commonly used to process sequence data.

[0067] In one or more embodiments of the present application, the network topology feature data refers to each node in the target network (i.e., the device participating in network communication) and the connection lines representing the links between each node, and each node is marked with the corresponding node state, and the connection lines are marked with information such as link state, etc. It can be understood that the node state can include node load, etc., and can also include network bandwidth, delay requirements, and link state, etc.

[0068] Step 200: Extract the resource allocation policy data corresponding to the target topological structure from the resource allocation policy data corresponding to each of the pre-stored predicted topological structures; among them, the resource allocation policy data corresponding to each of the predicted topological structures is pre-predicted by the SAT model according to the node feature vectors corresponding to each of the predicted topological structures and the service feature data; the node feature vectors corresponding to each of the predicted topological structures are pre-predicted by the GAT model according to the predicted topological structures corresponding to each of the predicted time points respectively.

[0069] In step 200, the resource allocation policy data mainly includes bandwidth management policy data, IP address management policy data, service quality (QoS) setting data, etc. The bandwidth management policy data may include network congestion control data, bandwidth priority division data, bandwidth dynamic allocation data, etc. The IP address management policy data may include IP address segment management data, dynamic IP address allocation data, etc. The service quality (QoS) setting data may include priority setting data and traffic monitoring tool adjustment data.

[0070] Step 300: Adjust the network resource allocation of the target network according to the resource allocation policy data corresponding to the target topology structure to achieve network fault recovery.

[0071] This application proposes an intelligent fast recovery strategy based on service levels, which solves the problems of slow response speed, unreasonable resource allocation, and lack of intelligence in network fault recovery through four main steps: predicting future network topologies, feature extraction, resource allocation optimization, and comprehensive decision-making.

[0072] In this application, not only advanced deep learning technologies such as LSTM, GAT, and SAT are used to solve network fault recovery problems, but also optimizations and innovations are made specifically for the specific challenges in network topology prediction and resource allocation. Although those skilled in the art may think of using LSTM for time series prediction, GAT for graph structure feature extraction, and SAT for resource allocation, due to the dynamic changes of the network fault scenario itself, topological complexity, and the timeliness requirements of resource allocation, these technologies face some technical obstacles and difficulties in practical applications.

[0073] For example, when dealing with network topology prediction, the traditional LSTM model often relies on global information and ignores the local dependencies between nodes. When facing complex network topologies and time-varying loads, this may lead to insufficient accuracy of the prediction results and be unable to accurately reflect the future network state. However, by introducing the graph attention network (GAT) in this application, the deficiency of LSTM in capturing local dependencies between nodes is made up for, ensuring accurate prediction of the network topology and being able to reflect the dynamic relationships between nodes and changes in the network structure.

[0074] In addition, the real-time adjustment of the network resource allocation policy also faces great challenges. In a dynamic network environment, the network state may change rapidly, resulting in the inability of traditional static rule-based resource allocation policies to adapt to the changes in the network state in a timely manner, and even may cause resource waste or network congestion. The self-attention mechanism (SAT) proposed in this application can intelligently adjust the resource allocation policy according to the real-time node feature vectors and network state, ensuring the continuous and stable operation of network services.

[0075] These innovative designs not only overcome the limitations of traditional technologies in specific network environments, but also improve the fault recovery efficiency and the reliability of network services by optimizing the model structure and strategy design. In summary, although technologies such as LSTM, GAT, and SAT are not novel in themselves, in the application scenario of this application, for the unique requirements of network fault recovery, these technologies have been customized and improved to enable them to overcome the obstacles of the existing technologies and provide more efficient and accurate solutions.

[0076] As can be seen from the above description, the network fault rapid recovery method provided by the embodiments of this application can effectively improve the response speed of network fault recovery, optimize resource allocation, reduce resource waste, and improve the automation level through intelligent decision-making, thereby solving the problems of slow response, unreasonable resource allocation, and lack of intelligence existing in traditional network fault recovery.

[0077] To ensure the accuracy of the prediction model and accurately predict the future network state, in a network fault rapid recovery method provided by the embodiments of this application, refer to Figure 2 , before step 100 in the network fault rapid recovery method, the following specific content is further included:

[0078] Step 010: Input the historical network state feature data and service feature data of the target network currently obtained into the LSTM model, so that the LSTM model outputs the predicted topological structures corresponding to the target network at each prediction time point respectively.

[0079] Specifically, a time series prediction model (such as the long short-term memory network LSTM) is used to predict the network topology for the next five moments (i.e., the prediction time points). The LSTM model is trained using historical network state data and service data to obtain a prediction model, which can capture the temporal law of network topology changes and then predict the network topology changes at future moments. The predicted network topologies at the five moments include information such as node status, link status, and resource availability in the network at each moment. This information will be input into the subsequent graph neural network (GAT) model as input to further extract the feature vectors of network nodes.

[0080] Step 020: Store the correspondence between each prediction time point and the predicted topological structure corresponding to each prediction time point locally.

[0081] That is to say, traditional methods usually rely on static historical data and simple statistical models, and it is difficult to accurately predict the future network state, especially in the case of frequent changes in network topology and load. By using the LSTM model in the embodiments of this application, the network topology at multiple future moments can be predicted more accurately, improving the accuracy and reliability of the prediction.

[0082] In order to extract the feature vectors of each node from the predicted network topology, capture the complex dependencies between nodes, and thus provide more comprehensive and rich feature information, in a network fault rapid recovery method provided in an embodiment of the present application, refer to Figure 2 , before step 100 in the network fault rapid recovery method, the following specific content is further included:

[0083] Step 030: Input the predicted topology structure and service feature data corresponding to the target network at each predicted time point output by the LSTM model into the GAT model, so that the GAT model correspondingly outputs the node feature vectors corresponding to the predicted topology structures at each of the predicted time points.

[0084] Specifically, the network topology predicted at each moment is respectively input into a preset graph neural network (GAT) model. Through graph convolution operations, the GAT model automatically learns the feature representations (i.e., node feature vectors) of each node under different topology structures. The feature vector of each node contains the state information of the node at the predicted moment, such as node load, network bandwidth, delay requirements, link status, etc. Through the GAT model, the feature representations of all nodes at each predicted time point can be obtained, and these feature vectors will provide an important basis for subsequent resource allocation and decision-making.

[0085] That is to say, traditional solutions often can only extract limited and fixed features and are difficult to capture the dynamic changes in the network. However, through the graph neural network (GAT) model, the feature vectors of each node can be extracted from the predicted network topology, and the complex dependencies between nodes can be captured, thereby providing more comprehensive and rich feature information.

[0086] In order to dynamically allocate resources according to the feature vectors of each node, better balance the network load, and improve resource utilization efficiency, in a network fault rapid recovery method provided in an embodiment of the present application, refer to Figure 2 , before step 100 in the network fault rapid recovery method, the following specific content is further included:

[0087] Step 040: Input the node feature vectors corresponding to the predicted topology structures output by the GAT model and the service feature data into the SAT model, so that the SAT model correspondingly outputs the resource allocation strategy data corresponding to the predicted topology structures.

[0088] Step 050: Store the corresponding relationship between the predicted topology structures corresponding to each predicted time point and the resource allocation strategy data corresponding to each of the predicted topology structures locally.

[0089] Specifically, based on the network state and node characteristics at each moment, a SAT model (a feasible resource allocation model or optimization model) is used for resource allocation. The SAT model considers factors such as the resource state of the current network, node load, bandwidth utilization, etc., and combines the network topology characteristics and service levels at each moment to calculate the optimal resource allocation strategy. The goal of resource allocation is to ensure that during the fault recovery period, network resources are reasonably scheduled and allocated, so as to maximize the satisfaction of service requirements and avoid resource waste or overload.

[0090] That is to say, traditional resource allocation methods usually rely on fixed rules or simple heuristic algorithms and are difficult to adapt to complex network environments. Through the self-attention mechanism (SAT) model, resource allocation can be dynamically performed according to the feature vectors of each node, better balancing the network load and improving resource utilization efficiency.

[0091] In order to further improve the response speed of network fault recovery and optimize resource allocation, in a network fault rapid recovery method provided in an embodiment of the present application, refer to Figure 2 , before step 010 in the network fault rapid recovery method, the following specific content is further included:

[0092] Step 001: Real-time obtain the network topology state data and service data respectively corresponding to the target network at each historical time point; wherein, the network topology state data includes the node state data respectively corresponding to each node in the target network and the link state data corresponding to the links between each node; the service data includes service level, service traffic, delay requirement data, and bandwidth demand data.

[0093] In step 001, assuming that the time point when obtaining the network topology state data and service data respectively corresponding to the target network at each historical time point is t1n, then each historical time point obtained includes this t1n and t11, t12, t13, etc. before t1n. That is to say, when obtaining the network topology state data and service data respectively corresponding to the target network at each historical time point, each historical time point actually includes the historical network state data before the time point t1n and the current network state data at the time point t1n.

[0094] In some embodiments of the present application, historical network state data and current network state data can be obtained, such as the network topology diagram of a communication network, where the network topology diagram includes each node in the communication network and link information between different nodes. And, obtain the current service level information to be allocated for the communication network.

[0095] Step 002: Clean the network topology status data and service data corresponding to each historical time point.

[0096] The above data can be obtained in real time through network facilities such as network monitoring devices, routers, and switches and transmitted to the data processing center. The data processing center cleans and organizes the collected data, removing noise data, outliers, and duplicate data to ensure the quality and integrity of the data.

[0097] Step 003: Extract features from the network topology status data corresponding to each historical time point after data cleaning to obtain corresponding historical network status feature data; and extract features from the service data after data cleaning to obtain corresponding service feature data.

[0098] Based on the cleaned data, perform feature engineering to extract useful features for subsequent prediction and resource allocation. The feature extraction module extracts spatio-temporal features such as network topology features, node status features, link bandwidth utilization, link latency, and node load from network status data (i.e., historical network status data and current network status data); and extracts service requirement features such as service level, service traffic, latency requirement, and bandwidth requirement from service data.

[0099] Step 004: Perform standardization and normalization processing on the historical network status feature data and service feature data.

[0100] Perform standardization processing and normalization processing on the extracted feature data (i.e., the historical network status feature data and service feature data) to ensure that each feature is under the same dimension, so that subsequent prediction models and decision algorithms can work efficiently and accurately.

[0101] To further improve the response speed of network fault recovery and optimize resource allocation, in a network fault fast recovery method provided in an embodiment of the present application, see Figure 2 , step 300 in the network fault fast recovery method specifically includes the following content:

[0102] Step 310: According to the resource allocation strategy data corresponding to the target topology structure, perform network topology adjustment, routing reconfiguration, and node load balancing processing on the target network.

[0103] To further improve the reliability and effectiveness of optimizing resource allocation, in a network fault fast recovery method provided in an embodiment of the present application, see Figure 2 , after step 310 in the network fault fast recovery method, it specifically further includes the following content:

[0104] Step 320: During the network fault recovery process of the target network, if it is detected that the predicted topology corresponding to the predicted time point matching the current latest time point previously output by the LSTM model has changed compared to the target topology, update the current target topology to the predicted topology corresponding to the predicted time point matching the current latest time point.

[0105] It can be understood that the predicted topology corresponding to the predicted time point matching the current latest time point is still predicted according to the GAT model and the SAT model.

[0106] Step 330: Obtain the resource allocation policy data corresponding to the updated target topology, and based on the resource allocation policy data corresponding to the updated target topology, re-perform network topology adjustment, routing reconfiguration, and node load balancing processing on the target network.

[0107] That is to say, the present application aims to provide an intelligent fast recovery strategy based on service levels, enabling the present application to provide a faster response speed, more reasonable resource allocation, and a higher level of intelligence in the face of network faults, thereby significantly improving the reliability and service quality of the network. First, collect historical network state data, service data, and current network state data; then clean and organize the data, and perform feature engineering to extract useful features for prediction and resource allocation; use a time series prediction model (LSTM) to predict the network topology for the next five moments; input the predicted network topologies for the five moments into a preset graph neural network GAT model, and the GAT model extracts the node features at each predicted moment to obtain the feature vectors of each node; for the network state at each moment, use the SAT model for resource allocation, and based on the resource allocation results at each moment, comprehensively consider the service level and the current network state to formulate a final fault recovery strategy. Traditional methods need to manually adjust rules or parameters in the face of network topology and load changes, with a slow response speed and prone to errors. By combining time series prediction and graph neural networks, the present application can automatically adapt to network changes, make a quick response, and has higher adaptability and flexibility. Traditional fault recovery methods usually rely on preset fixed strategies and are difficult to dynamically adjust according to the current network state and service requirements. However, the present application can formulate a final fault recovery strategy by comprehensively considering the service level and the current network state, and can more efficiently restore network services and reduce the fault impact time. Traditional solutions have a high computational complexity and are difficult to scale when dealing with large-scale networks. The present application can effectively process large-scale network data by using deep learning models, has good scalability, and is applicable to network environments of various scales.

[0108] In summary, the technical solution of this application combines time series prediction, graph neural network, and self-attention mechanism to achieve more accurate network state prediction, more comprehensive feature extraction, more intelligent resource allocation, and more efficient fault recovery, significantly improving the effect of network management and optimization.

[0109] To further illustrate the above embodiments, this application also provides a specific application example of a network fault rapid recovery method. The intelligent rapid recovery strategy based on service level is a solution that emerged under such a background. It aims to provide corresponding recovery support for high-level services preferentially through intelligent means according to the importance and impact degree of services, so as to restore the operation of critical services in a relatively short time and reduce the losses caused by faults. The application example of this application combines the application of graph neural networks (GNNs) in the intelligent rapid recovery strategy based on service level, which can further improve the intelligent level of the recovery process. Graph neural network is a powerful tool for processing graph-structured data and can capture complex dependency relationships between nodes. It can be used to construct and analyze complex network topologies, and then optimize the recovery strategy. The application example of this application proposes an intelligent rapid recovery strategy based on service level, aiming to effectively solve the above problems through four main steps: predicting the future network topology, feature extraction, resource allocation optimization, and comprehensive decision-making. This strategy improves the response speed and resource utilization efficiency of the network fault recovery process by introducing intelligent prediction and optimization mechanisms, ensuring that the requirements of different service levels are reasonably met.

[0110] The technical key point of the application example of this application is to use a long short-term memory network (LSTM) time series prediction model to predict the network topology at multiple future moments, ensuring the accuracy of the prediction model and being able to accurately predict the future network state; using a graph attention network (GAT) model to extract the feature vectors of each node from the predicted network topology, ensuring that the extracted feature vectors can comprehensively reflect the state of the nodes and the dependency relationships between the nodes; using a self-attention mechanism (SAT) model to allocate resources according to the feature vectors of each node, and adjusting the resource allocation strategy in real time according to the dynamic changes of the network state. According to the resource allocation results at each moment, considering the service level and the current network state comprehensively, a final fault recovery strategy is formulated to ensure that the fault recovery strategy can efficiently and quickly restore network services and reduce the fault impact time.

[0111] The protected point of the application example of this application is to use LSTM to predict the most likely network topology situation in the next five moments. The purpose of doing this is to enable quick recovery in case of failures. Then, it is input into GAT because GAT can process different types of graph data, enhance the expressive ability, and can perform better in various different scenarios, especially when dealing with complex, heterogeneous or noisy graph data. Therefore, in the application example of this application, GAT is used to extract node features, and then the SAT model is used for resource allocation.

[0112] See Figure 3 , the following is the specific implementation manner of the application example of this application:

[0113] Step 1: Data collection and cleaning

[0114] First of all, the system collects historical network state data, business data, and current network state data. The historical network state data includes information such as network topology structure, node status, and link status within a past period of time; the business data includes different business types running in the network, business levels, and demands for network resources (i.e., network resource demand data); the current network state data includes real-time network topology structure, node status such as node load, and link status information such as link delay.

[0115] The data collection module obtains the above data in real time through network facilities such as network monitoring devices, routers, and switches and transmits it to the data processing center. The data processing center cleans and organizes the collected data, removes noise data, outliers, and duplicate data to ensure the quality and integrity of the data.

[0116] Step 2: Feature engineering and data preprocessing

[0117] Based on the cleaned data, feature engineering is carried out to extract useful features for subsequent prediction and resource allocation. The feature extraction module extracts spatio-temporal features such as network topology features, node status features, link bandwidth utilization, link delay, and node load from the network state data (i.e., historical network state data and current network state data); and extracts business demand features such as business level, business traffic, delay requirements, and bandwidth demands from the business data.

[0118] The data preprocessing module standardizes and normalizes the extracted feature data to ensure that each feature is in the same dimension so that the subsequent prediction model and decision algorithm can work efficiently and accurately.

[0119] Step 3: Future network topology prediction

[0120] After feature extraction is completed, a time series prediction model (such as the long short-term memory network LSTM) is used to predict the network topology for the next five time instants. The LSTM model is trained using historical network state data and service data to obtain a prediction model, which can capture the temporal pattern of network topology changes and then predict the network topology changes at future time instants.

[0121] The predicted network topologies for the five time instants include information such as the node state, link state, and resource availability in the network at each time instant. This information will be used as input and fed into the subsequent graph neural network (GAT) model to further extract the feature vectors of network nodes.

[0122] Specifically, the Recurrent Neural Network (RNN) is a type of artificial neural network, and LSTM is its variant. The biggest difference between LSTM and the traditional RNN lies in the structure of its hidden layer module. The LSTM hidden layer module is also called the memory module, and its specific structure is as Figure 4 shown. The memory module consists of a storage unit and three gates, which are called the input gate, output gate, and forget gate respectively, and they control the reading, writing, and resetting of data in the storage unit. To achieve the memory of input information, LSTM deletes or updates the information in the storage unit of the memory module through the input gate, output gate, and forget gate. Figure 4 The four units (including three sigmas and one tanh) at the bottom layer in the middle part of Figure 4 are the activation functions inside the memory module, and the other parts represent the basic operations of vectors.

[0123] See Figure 5 , in each LSTM unit, each box represents a neural network layer (Neural NetworkLayer), which consists of weights, biases, and activation functions; each circle represents an element-level operation (PointwiseOperation); the arrow represents the vector flow direction; the intersecting arrows represent the vector concatenation (Vector Transfer); the forked arrows represent the vector replication (Concatenate).

[0124] The core part of LSTM is in Figure 4 the uppermost part similar to a conveyor belt in Figure 6 ), which is generally called the cell state, and it exists throughout the entire chain system of LSTM.

[0125] Figure 6It can be expressed as the following formula:

[0126]

[0127] where f t It is called the forget gate, which means C t-1 Which features of are used to calculate C t .f t is a vector, each element of which is in the range [0,1]. Sigmoid is usually used as an activation function, and the output of sigmoid is a value in the range [0,1]. However, when observing a trained LSTM, it is found that most of the gate values are very close to 0 or 1, and the rest are very rare. It is the most important gate mechanism of LSTM, indicating f t The unit multiplication relationship between and, W f , W i , W C , W o It is represented as the corresponding weight coefficient matrix.

[0128] Specifically, Figure 7 It can be expressed as:

[0129] f t =σ(W f *[h t-1 ,x t ]+b f )

[0130] like Figure 8 and Figure 9 As shown, Represents the updated value of the unit state, which is determined by the input data x t and hidden node h t-1 The activation function of the unit state update value obtained through a neural network layer is usually tanh. t It is called the input gate, which is the same as f t It is also a vector with elements between the interval [0,1], and is also composed of x t and h t-1 Calculated by the sigmoid activation function.

[0131] Specifically, Figure 8 It can be expressed as:

[0132] i t =σ(W i *[h t-1 ,x t ]+b i )

[0133]

[0134] Finally, to calculate the predicted value and generate the complete input for the next time slice, it is necessary to calculate the output h of the hidden node t , as Figure 10 shown

[0135] o t = σ(W o * [h t-1 , x t + b o )

[0136] h t = o t * tanh(C t )

[0137] h t is obtained from the output gate and the cell state, where the calculation method of o t is the same as that of f t . LSTM is a powerful tool for processing time series data and has great advantages for time-related tasks such as traffic prediction. Through appropriate data preprocessing and network tuning, LSTM can better capture the patterns and regularities in time series data, improve the accuracy of prediction, and effectively capture long-term dependencies.

[0138] Step 4: Graph neural network feature extraction

[0139] Input the network topology at each moment predicted in Step 3 into a preset graph neural network (GAT) model respectively. The GAT model automatically learns the feature representation (i.e., the node feature vector) of each node under different topological structures through graph convolution operations. The feature vector of each node contains the state information of the node at the prediction moment, such as node load, network bandwidth, delay requirement, link state, etc.

[0140] Through the GAT model, the feature representations of all nodes at each prediction moment can be obtained, and these feature vectors will provide an important basis for subsequent resource allocation and decision-making.

[0141] Specifically, after predicting the network topology for the next five moments, input the network topologies at the five predicted moments into the preset graph neural network GAT model respectively. The GAT model extracts the node features at each prediction moment to obtain the feature vector of each node. The feature vector can be used to evaluate the importance and state of the node at a specific moment.

[0142] The GAT model refers to Graph Attention Networks, which is a model for performing machine learning operations on graph-structured data. It belongs to a type of Graph Neural Networks (GNNs) and is mainly used to process data structures in non-Euclidean spaces, such as social networks. GAT aggregates neighbor nodes through a self-attention mechanism, achieving an adaptive allocation of weights for different neighbors. In traditional graph convolutional networks, the update of node features depends on a pre-constructed graph structure and usually requires complex matrix operations, such as the Laplacian matrix. In contrast, GAT does not need to use these matrices but updates node features through the first-order neighbors of the nodes, thus realizing a more efficient graph neural network.

[0143] The core of GAT is the use of a masked self-attention layer, which allows each node to assign different weights to its neighbors according to the features of its neighbor nodes. This means that each node can be weighted according to the importance of its neighbor nodes when updating its features, thus better capturing the complex relationships between nodes. Through this attention mechanism, GAT can more flexibly model graph data and is applicable to a wider range of graph structures and tasks. Specifically, the calculation steps of GAT include:

[0144] (1) Calculate the attention coefficient: Calculate the attention scores between nodes through a single-layer feedforward neural network (usually with a LeakyReLU activation function), and then use the softmax function to normalize among neighbor nodes.

[0145] In the process of calculating the attention coefficient, the goal of GAT is to assign a weight to the connection between each node and its neighbors, and this weight reflects the importance of the neighbor nodes to the central node. The specific steps are as follows:

[0146] Feature transformation: See Figure 11 , first, map the feature vector of each node to a new feature space through a learnable linear transformation W. This step can be achieved through matrix multiplication, that is, Wh i , where h i is the feature vector of node i, and W is the weight matrix.

[0147] Attention score calculation: Next, calculate the attention score between node i and its neighbor node j. This is usually achieved through a single-layer feedforward neural network. The network accepts the concatenation of the feature vectors of nodes i and j as input and outputs a scalar. The formula can be expressed as Here is a parameterized function, usually a single-layer feedforward neural network.

[0148] Normalization: Finally, normalize the attention scores e ij among all neighbors of node i to form the attention coefficients α ij . Normalization is usually implemented using the softmax function to ensure that the sum of the attention coefficients of all neighbors equals 1. The formula is as follows:

[0149]

[0150] where N i is the set of neighbors of node i.

[0151] (2) Weighted summation (aggregate): After obtaining the attention coefficients, the next step is to perform a weighted summation of the feature vectors of the neighbor nodes to update the feature representation of the central node. The specific steps are as follows:

[0152] Use the calculated attention coefficients a ij to weight the feature vector h j of neighbor node j to obtain the new feature representation of central node i. The formula is shown as follows: where σ is the activation function used to increase non-linearity.

[0153] Step Five: Resource Allocation and Fault Recovery Decision

[0154] Based on the network state and node features at each moment, use the SAT model (a feasible resource allocation model or optimization model) for resource allocation. The SAT model considers factors such as the current network's resource state, node load, bandwidth utilization, etc., and combines the network topology features and service levels at each moment to calculate the optimal resource allocation strategy. The goal of resource allocation is to ensure that during fault recovery, network resources are reasonably scheduled and allocated, thus maximizing the satisfaction of service requirements and avoiding resource waste or overload.

[0155] Specifically, use the method mentioned above to construct the GAT model and train the model to learn the feature vectors of each node. After obtaining the feature vectors of each node, for the network state at each moment, use the SAT model for resource allocation. Using the Self-Attention Mechanism (SAT) for resource allocation can play an important role in multiple fields such as network management, task scheduling, and logistics optimization.

[0156] The SAT model can capture the dependencies between nodes and make optimized decisions based on these relationships. The following is a detailed step description showing how to use the SAT model for resource allocation: First, collect network status data, including historical data and real-time data. Historical data is the data collected on the past network operating status, including node characteristics (such as bandwidth utilization, latency, packet loss rate, etc.) and connection relationships. Real-time data is to obtain the current network status data for real-time resource allocation. After obtaining these data, perform data preprocessing on the data, check and clean the outliers or missing values in the data. Extract useful features from the original data, such as the load of nodes, the trend of throughput change, etc. Transform the data to the same scale for subsequent processing. Secondly, construct the SAT model. The input layer of the model is the feature vector of the receiving nodes. The self-attention layer is mainly responsible for calculating the attention distribution between nodes. The fully connected layer generates the resource allocation scheme according to the attention distribution. Through the above steps, the self-attention mechanism (SAT) model can be used to effectively perform resource allocation, improving network performance and resource utilization efficiency.

[0157] When performing resource allocation, the system also comprehensively considers the level requirements of different services. For example, for high-priority services, the system will preferentially allocate more resources to ensure the guarantee of their network performance; while for low-priority services, the system will make reasonable scheduling according to the remaining resources.

[0158] According to the resource allocation results at each moment, comprehensively considering the service level and the current network status, formulate the final fault recovery strategy. This strategy can not only quickly respond to network faults, but also ensure the rationality and intelligence of resource allocation during the recovery process.

[0159] Among them, the fault recovery process:

[0160] Once a network fault occurs, the system will automatically execute the fault recovery process based on the predicted network topology changes and resource allocation scheme. The recovery process includes the following steps:

[0161] (1) Quickly identify the network topology after the fault: Based on the predicted network topology changes, quickly identify the possible network topologies after the fault.

[0162] (2) Adjust resource allocation: According to the changes in faulty nodes and links, adjust the allocation of network resources to ensure the continuity and efficiency of services during network recovery. Perform necessary network topology adjustments, routing reconfigurations, node load balancing, etc. operations to minimize the impact of the fault on services.

[0163] (3) Dynamic optimization: During the recovery process, the system dynamically adjusts the resource allocation strategy according to the changes in the real-time network status to ensure that the recovery process continues to be efficient and avoid new bottlenecks or load overloading.

[0164] Through four steps of predicting future network topology, feature extraction, resource allocation optimization, and comprehensive decision-making, the application example of this application proposes an intelligent fast recovery strategy based on service levels. This strategy can effectively improve the response speed of network fault recovery, optimize resource allocation, reduce resource waste, and improve the automation level of the system through intelligent decision-making, thus solving the problems of slow response, unreasonable resource allocation, and lack of intelligence in traditional network fault recovery.

[0165] At the software level, this application also provides a network fault fast recovery device for executing all or part of the content in the network fault fast recovery method. Refer to Figure 12 The network fault fast recovery device specifically includes the following content:

[0166] The prediction topology matching module 10 is used to, if it detects that a network fault occurs in the target network currently, select, from the predicted topology structures corresponding to the target network at each predicted time point pre-obtained, a predicted topology structure corresponding to a predicted time point that matches the current time point as the target topology structure; wherein, the predicted topology structures corresponding to each predicted time point are pre-predicted by the LSTM model according to the historical network state feature data and service feature data of the target network.

[0167] The allocation strategy extraction module 20 is used to extract the resource allocation strategy data corresponding to the target topology structure from the resource allocation strategy data corresponding to each of the pre-stored predicted topology structures; wherein, the resource allocation strategy data corresponding to each of the predicted topology structures are pre-predicted by the SAT model according to the node feature vectors corresponding to each of the predicted topology structures and the service feature data; the node feature vectors corresponding to each of the predicted topology structures are pre-predicted by the GAT model according to the predicted topology structures corresponding to each predicted time point respectively.

[0168] The fault recovery module 30 is used to adjust the network resource allocation of the target network according to the resource allocation strategy data corresponding to the target topology structure to achieve network fault recovery. The embodiment of the network fault fast recovery device provided by this application can specifically be used to execute the processing flow of the embodiment of the network fault fast recovery method in the above embodiment, and its functions will not be elaborated here. Reference can be made to the detailed description of the embodiment of the network fault fast recovery method above.

[0169] The part of the network fault rapid recovery device for network fault rapid recovery can be completed in the client device. Specifically, it can be selected according to the processing capacity of the client device and the limitations of the user usage scenario, etc. This application does not make any limitations in this regard. If all operations are completed in the client device, the client device may further include a processor for specific processing of network fault rapid recovery.

[0170] The above-mentioned client device may have a communication module (i.e., communication unit), which can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side, and in other implementation scenarios, it may also include a server of an intermediate platform, such as a server of a third-party server platform having a communication link with the task scheduling center server. The server may include a single computer device, or a server cluster composed of multiple servers, or a server structure of a distributed device.

[0171] Any suitable network protocol can be used for communication between the above-mentioned server and the client device side, including network protocols not yet developed on the filing date of this application. The network protocol may, for example, include TCP / IP protocol, UDP / IP protocol, HTTP protocol, HTTPS protocol, etc. Of course, the network protocol may also, for example, include RPC protocol (Remote Procedure Call Protocol) and REST protocol (Representational State Transfer) used on top of the above-mentioned protocols.

[0172] From the above description, it can be seen that the network fault rapid recovery device provided by the embodiments of this application can automatically change the eye features of the characters in the video data, and by adopting a dedicated target eye feature change model for the target character under the target change requirements, it can effectively improve the pertinence and effect of the change. On this basis, by adopting the method of transfer learning, it can also effectively improve the efficiency and automation degree of the change process, effectively reduce the labor cost and time cost consumed in the later processing, and can effectively improve the user experience of video viewers, thereby meeting the change requirements of the eye movement features of characters applicable to scenarios such as movie settings.

[0173] The embodiments of this application also provide an electronic device, which may include a processor, a memory, a receiver, and a transmitter. The processor is used to execute the network fault rapid recovery method mentioned in the above embodiments. The processor and the memory may be connected through a bus or other means. Taking the connection through the bus as an example, the receiver can be connected to the processor and the memory in a wired or wireless manner.

[0174] The processor may be a Central Processing Unit (CPU). The processor may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., such as chips, or combinations of the above types of chips.

[0175] As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the network fault fast recovery method in the embodiments of the present application. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory, that is, implements the network fault fast recovery method in the above method embodiments.

[0176] The memory may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created by the processor, etc. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory may optionally include a memory remotely provided relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above networks include but are not limited to the Internet, enterprise intranets, local area networks, mobile communication networks, and combinations thereof.

[0177] The one or more modules are stored in the memory and, when executed by the processor, execute the network fault fast recovery method in the embodiments.

[0178] In some embodiments of the present application, the user equipment may include a processor, a memory, and a transceiver unit. The transceiver unit may include a receiver and a transmitter. The processor, the memory, the receiver, and the transmitter may be connected through a bus system. The memory is used to store computer instructions, and the processor is used to execute the computer instructions stored in the memory to control the transceiver unit to transmit and receive signals.

[0179] As an implementation manner, the functions of the receiver and the transmitter in the present application can be considered to be implemented through a transceiver circuit or a dedicated transceiver chip, and the processor can be considered to be implemented through a dedicated processing chip, a processing circuit, or a general-purpose chip.

[0180] As another implementation, it is possible to consider implementing the server provided in the embodiments of the present application by using a general-purpose computer. That is, the program codes for implementing the functions of the processor, receiver, and transmitter are stored in the memory, and the general-purpose processor implements the functions of the processor, receiver, and transmitter by executing the codes in the memory.

[0181] The embodiments of the present application further provide a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the foregoing method for quickly recovering from a network failure are implemented. The computer-readable storage medium may be a tangible storage medium, such as a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disks, removable storage disks, CD-ROMs, or any other form of storage medium known in the art.

[0182] The embodiments of the present application further provide a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the foregoing method for quickly recovering from a network failure are implemented.

[0183] Those of ordinary skill in the art should understand that the various exemplary components, systems, and methods described in connection with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Specifically, whether to execute in hardware or software depends on the specific application and design constraints of the technical solution. A professional technician can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a functional card, and so on. When implemented in software, the elements of the present application are programs or code segments used to execute the required tasks. The program or code segment can be stored in a machine-readable medium or transmitted through a data signal carried in a carrier wave on a transmission medium or a communication link.

[0184] It should be clear that the present application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, the detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order between steps after understanding the spirit of the present application.

[0185] In this application, features described and / or illustrated for one embodiment can be used in the same or a similar manner in one or more other embodiments, and / or combined with the features of other embodiments or replace the features of other embodiments.

[0186] The foregoing are only the preferred embodiments of this application and are not intended to limit this application. For those skilled in the art, various changes and modifications can be made to the embodiments of this application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included within the protection scope of this application.

Claims

1. A method for rapid recovery of network failures, characterized in that, Including: If it is detected that a network failure has occurred in the target network currently, from the predicted topology structures corresponding to the target network at each predicted time point obtained in advance, select the predicted topology structure corresponding to one of the predicted time points that matches the current time point as the target topology structure; wherein, the predicted topology structures corresponding to each predicted time point are respectively predicted by the LSTM model according to the historical network state feature data and service feature data of the target network; Extract the resource allocation policy data corresponding to the target topology structure from the resource allocation policy data corresponding to each of the predicted topology structures stored in advance; wherein, the resource allocation policy data corresponding to each of the predicted topology structures are respectively predicted by the SAT model according to the node feature vectors corresponding to each of the predicted topology structures and the service feature data; the node feature vectors corresponding to each of the predicted topology structures are respectively predicted by the GAT model according to the predicted topology structures corresponding to each predicted time point; Adjust the network resource allocation of the target network according to the resource allocation policy data corresponding to the target topology structure to achieve network failure recovery.

2. The network fault rapid recovery method according to claim 1, wherein Before the step of "If it is detected that a network failure has occurred in the target network currently, from the predicted topology structures corresponding to the target network at each predicted time point obtained in advance, select the predicted topology structure corresponding to one of the predicted time points that matches the current time point as the target topology structure", it further includes: Input the currently obtained historical network state feature data and service feature data of the target network into the LSTM model, so that the LSTM model outputs the predicted topology structures corresponding to the target network at each predicted time point respectively; Store the correspondence between each predicted time point and the predicted topology structure corresponding to each predicted time point locally.

3. The network fault fast recovery method according to claim 1, characterized in that Before the step of "If it is detected that a network failure has occurred in the target network currently, from the predicted topology structures corresponding to the target network at each predicted time point obtained in advance, select the predicted topology structure corresponding to one of the predicted time points that matches the current time point as the target topology structure", it further includes: Input the predicted topology structures corresponding to the target network at each predicted time point output by the LSTM model and the service feature data into the GAT model, so that the GAT model outputs the node feature vectors corresponding to the predicted topology structures at each predicted time point respectively.

4. The network fault fast recovery method according to claim 1, wherein Before the step of "If it is detected that a network failure has occurred in the target network currently, from the predicted topology structures corresponding to the target network at each predicted time point obtained in advance, select the predicted topology structure corresponding to one of the predicted time points that matches the current time point as the target topology structure", it further includes: Input the node feature vectors corresponding to the predicted topology structures output by the GAT model and the service feature data into the SAT model, so that the SAT model outputs the resource allocation policy data corresponding to the predicted topology structures respectively. Store the correspondence between the predicted topology structures corresponding to the respective predicted time points and the resource allocation policy data corresponding to the respective predicted topology structures locally.

5. The network fault fast recovery method according to claim 2, wherein The inputting the historical network state feature data and service feature data of the target network currently obtained into the LSTM model includes: Obtain in real time the network topology state data and service data respectively corresponding to the target network at each historical time point; wherein, the network topology state data includes the node state data respectively corresponding to each node in the target network and the link state data corresponding to the links between each node; the service data includes service level, service traffic, latency requirement data, and bandwidth demand data; Perform data cleaning on the network topology state data and service data respectively corresponding to each historical time point. Extract features from the network topology state data respectively corresponding to each historical time point after data cleaning to obtain corresponding historical network state feature data; and extract features from the service data after data cleaning to obtain corresponding service feature data. Perform standardization and normalization processing on the historical network state feature data and service feature data.

6. The network fault fast recovery method according to claim 1, characterized in that The adjusting the network resource allocation of the target network according to the resource allocation policy data corresponding to the target topology structure to achieve network fault recovery includes: According to the resource allocation policy data corresponding to the target topology structure, perform network topology adjustment, routing reconfiguration, and node load balancing processing on the target network.

7. The network fault rapid recovery method according to claim 6, wherein The adjusting the network resource allocation of the target network according to the resource allocation policy data corresponding to the target topology structure to achieve network fault recovery further includes: During the network fault recovery process of the target network, if it is detected that the predicted topology structure corresponding to the predicted time point matching the current latest time point pre-output by the LSTM model has changed compared with the target topology structure, update the current target topology structure to the predicted topology structure corresponding to the predicted time point matching the current latest time point; Obtain the resource allocation policy data corresponding to the updated target topology structure, and based on the resource allocation policy data corresponding to the updated target topology structure, re-perform network topology adjustment, routing reconfiguration, and node load balancing processing on the target network.

8. A network fault rapid recovery device, characterized in that Includes: A predicted topology matching module, configured to, if it is detected that a network fault occurs in the target network currently, select, from the predicted topology structures respectively corresponding to the target network at each predicted time point pre-obtained, the predicted topology structure corresponding to one predicted time point matching the current time point as the target topology structure; wherein, the predicted topology structures respectively corresponding to each predicted time point are pre-predicted by the LSTM model according to the historical network state feature data and service feature data of the target network; The allocation strategy extraction module is used to extract the resource allocation strategy data corresponding to the target topology structure from the resource allocation strategy data corresponding to each of the pre-stored prediction topology structures; wherein, the resource allocation strategy data corresponding to each of the prediction topology structures are pre-predicted by the SAT model according to the node feature vectors corresponding to each of the prediction topology structures and the service feature data; the node feature vectors corresponding to each of the prediction topology structures are pre-predicted by the GAT model according to each of the prediction topology structures corresponding to each prediction time point respectively; The fault recovery module is used to adjust the network resource allocation of the target network according to the resource allocation strategy data corresponding to the target topology structure to achieve network fault recovery.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the network fault rapid recovery method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the network fault rapid recovery method according to any one of claims 1 to 7.

Images